Applications of CST to modelling human interaction with EM fields: a metrological perspective. Benjamin Loader Date 6 Aug 2013
Contents 1. Introduction 2. Accuracy of models. 3. Some applications at NPL - designing and characterising exposure systems -testing safety of medical implants during MRI. - on-body and implanted antenna characterisation. 4. Conclusions
1. Introduction - computer used for simulations. Computer: Dell Precision T5500 - Dual Xeon quad core E5506 processors (2.13 GHz) - 48 GB RAM - 2 X 300 GB drives (10,000 rpm) (too small!) - 256 MB NVIDIA graphics card - 64 bit operating system.
Software used. CST Microwave Studio is used for most of the electromagnetic simulations at NPL. It was selected on the basis of accuracy for a series of benchmark problems. It is also very user friendly. - Solvers: Time domain, Eigen mode, Thermal, Circuit Simulator - 3-D import (CAD and Voxel models) - -Optimization and parameter sweep. Human voxel models: - HUGO, IT IS Virtual family and Virtual classroom.
2. Accuracy of human EM simulations Exact p.p.m. % db Random numbers Device related uncertainties (what you make is not usually what you modelled!) People are not standard (for what % of the population is your model valid?) Material parameters are not exact (material may be inhomogeneous, anisotropic and different in-vivo and in-vitro) Post-processing method (e.g. cubic volume, contiguous etc.) Solver accuracy (including boundary conditions) Standardisation of models improves agreement between groups but not necessarily accuracy!
How good is the input data? Published tissue properties are not exact. Tissues may be anisotropic, non-uniform may vary with age, and different in-vivo and in-vitro. Also, tissue structure mean results will depend on scale. Graphs for dielectric properties of long bone* * http://www.mthr.org.uk/research_projects/documents/rum3finalreport.pdf
Solid silicone rubber phantom Indicated field reading (Vm -1 ) 6 5 4 3 2 1 0 On-person (me again!) 0 90 180 270 360 0.1 GHz 0.2 GHz 0.9 GHz 1.85 GHz 2.44 GHz 5.2 GHz Azimuth angle (degrees) Indicated field reading (Vm -1 ) 6 5 4 3 2 1 0 On-rubber man 0 90 180 270 360 Azimuth angle (degrees) 0.1 GHz 0.2 GHz 0.9 GHz 1.85 GHz 2.44 GHz 5.2 GHz We would like to model it accurately to better understand the measurement results.
But the material is inhomgeneous and anisotropic! The standard deviations of the 16 measurements are: ε = 2.5 σ = 0.25 S/m!
And complex permittivity changes with frequency This requires dispersive material properties for the tissues or phantom for 3.1 to 10.6 GHz (i.e. the complex permittivity changes with frequency) UWB antenna
Dispersive tissue properties 4-Cole-Cole With ionic conductivity term (Gabriel) e.g. muscle ε 50 7000 := 1.2 10 6 τ 2.5 10 7 := 7.234 10 12 353.678 10 9 318.310 10 6 2.274 10 3 s α := 0.1 0.1 0.1 0 σ r := 0.2 S m ε f := 4.0
First order Debye with ionic conductivity term. This gives a good fit to the Gabriel data for the frequency range 1 to 11 GHz ε ( f ) jσ s ε s ε = ε + ωε 1 + jωτ τ = 1 2πF 0 R CST has 1 st and 2 nd order Debye models, but does not include the ionic conductivity term!!
Solution. Use the Macro- DefineHumanMaterialProperties, which generates the permittivity data table using the 4-Cole-Cole and then attempts an n th -order fit to the data. We generate a materials property table with 501 frequencies over the range 0.1 GHz to 40 GHz using the 1 st order Debye with ionic conductivity term, then fit a 2 nd order relaxation equation to this data. This was found to give very close agreement for the material properties over 1 to 11 GHz.
Voxel models of humans for computer simulation. HUGO (visible man project) is very large adult male (+95 percentile) and this is not suitable for all applications, and some tissues are missing, due to the method of data collection e.g. CSF in the head. IT IS virtual family and virtual classroom human voxel models can be imported into CST. Data sets are available for a range of people, and they are based on MRI scan data.
IT IS Virtual population models Name Sex Age Height Weight BMI [years] [m] [kg] [kg/m²] Roberta female 5 1.09 17.8 14.9 Thelonious male 6 1.17 19.3 14.0 Eartha female 8 1.36 30.7 16.7 Dizzie male 8 1.40 26.0 13.4 Billie female 11 1.47 35.4 16.5 Louis male 14 1.69 50.4 17.7 Ella female 26 1.63 58.7 22.0 Duke male 34 1.77 72.4 23.1 http://www.itis.ethz.ch/services/ anatomical-models/virtual-population/
Importing IT IS models to CST Models are supplied as *.SAT files. The voxel model for the required region and resolution is generated using SEMCAD-X ( supplied with the models). This outputs the *.raw file. The text file *.vox must be edited to reference the new model. You enter the number of voxels in each direction (x,y,z) for the data set. The format of the materials file for the IT IS models is different to that for CST import, so a compatible materials file must be made. (This can be done using data for all tissues from the Gabriel data base at the required frequency. http://niremf.ifac.cnr.it/tissprop/)
Example of the *.vox file for CST Version] 1.0 [Material] //f [MHz] filename 2400 Material_2400.txt [Background] 16 [Voxel] //type nx ny nz dx[mm] dy[mm] dz[mm] offset filename char 435 252 400 1 1 1 148 Dizzy.raw
Example of materials file *.txt for CST # Version 2009 ### // Frequency: 2450 MHz // Tissue Num Eps Mue Kappa Rho // [s/m] [kg/m^3] // Marrow 1 5.296872 1.0 0.095031 1030 FatTissue 2 5.280096 1.0 0.104517 1100 Bones 3 11.381223 1.0 0.394277 1850
- CST model showed ferrites were needed to reduce current induced on cable shield Block of material = 2/3 rds muscle Feed cables Ferrites Handset
- and gave an estimate for the size of amplifier required for the system. The model was not relied on to set the exposure level, rather the system was calibrated using a SAR measurement system and phantom to determine the required input power.
Which E-field orientation yields highest exposure? 1.25W/kg @ 100V/m 0.022 W/kg @100V/m 0.05 W/kg @ 100V/m
Testing the safety of implant during MRI Shielded bird cage resonator system (64 MHz) CST model showing power absorption in the legs
Some examples of hip-joint systems.
The heating of surrounding tissues during MRI can be a problem. Temperature Change (C) Fig. a 1.25 1 0.75 0.5 0.25 0 0 500 1000 Time (s) Temperature Change (C) Fig. b 1.25 1 0.75 0.5 0.25 0 0 500 1000 Time (s) Temperature Change (C) Fig. c 1.25 1 0.75 0.5 0.25 0 0 500 1000 Time (s)
Multi-physical modelling is use to: 1. Identify positions producing highest temperature rise for the tests 2. To determine the test position in the ASTM box phantom that gives similar local to whole body SAR as for the person. 3. To link the in-vitro (phantom) results to in-vivo (human) result. This is an example were it would be better to rely on the computer simulation to assess the safety of a passive implant, provided standard coil models are supplied.
Body area network simulation at 2.4 GHz HUGO multitissue anatomic model (1mm resolution) Computed parameters - insertion losses between the antennas - Far-field patterns and antenna efficiencies. - Model repeated with HOMOGENEOUS properties Monopole antenna above metal disc (x19), excited by discrete port
Far field patterns at 2.4 GHz. Heterogeneous Homogeneous 2/3 muscle
Simulation resultsantenna efficiency Radiated efficiency (linear) 0.8 0.6 0.4 0.2 0.0 heterogeneous 2/3 muscle "dry skin" 0 10 20 30 40 Antenna to body separation (mm)
Simulation resultson-body channel insertion losses. Sn,1 in db 0-20 -40-60 -80-100 Antennas on the back heterogeneous 2/3 muscle "dry skin" Antennas on the front 0 5 10 15 20 Receive port (n)
Using the simulation we are able to demonstrate that homogeneous phantom having the dielectric properties of dry skin or 2/3 muscle will give a good approximation for the human body for on-body channel characterisation measurements at 2.4 GHz. However, the antenna efficiency is of the order of 25-35% higher in the case of the homogeneous phantoms, so that some correction to the measurement results may be required. It would be very time consuming to demonstrate this by measurements for so many transmitter positions.
Modelling a simplified phantom shape -parameter sweep to see the effect of tissue layering Monopole antennas with discrete port (There is an antenna located on the lower face ) Computed parameters: S- parameters, far-fields, radiated efficiency Phantom can be layered or homogeneous
Current density
Simulation results. Description Fat (mm) Resonant at (GHz) S11 (db) S21 (db) S31 (db) Rad. eff. (linear). Homogeneous 2/3 muscle n/a 2.375-15.2-26.1-61.3 0.62 Homogeneous 3/5 muscle n/a 2.413-16.5-26.3-49.2 0.61 Homogeneous dry skin n/a 2.395-15.5-26.5-54.7 0.59 Layered Wet skin 5 2.387-12.8-28.2-51.1 0.56 Layered Dry skin 5 2.393-12.8-28.7-50.7 0.56 Layered Dry skin 10 2.386-12.2-27.7-52.2 0.65 Layered Dry skin 15 2.379-12.5-26.5-50.5 0.73 Layered Dry skin 20 2.376-13.0-25.7-47.8 0.76 Layered Dry skin 25 2.378-13.6-25.4-43.8 0.74 Layered Dry skin 30 2.378-13.7-25.8-36.5 0.66
Effect of fat thickness layer on radiated efficiency -1 Radiation efficiency db -1.5-2 -2.5-3 5 10 15 20 25 30 Fat layer thickness (mm)
4. Conclusions You cannot usual measure EM field in-vivo. Human voxel models are much more realistic and detailed than measurement phantoms. 3-D field distributions can be visualised, to give a better understanding of interaction of fields with devices and anatomy, which is good for designing things. You can model more cases than you can measure! Multi-physical models can give tissue temperatures, and temperature distributions, which is usually the biological end point. You cannot easily obtain temperature distributions in the phantom.
But Models are only as good as the input data and physical devices may be different from what was modelled. The model will not tell you whether the device is working correctly. Models are usually complex with many variables, and the results from different groups may not agree well. The range of validity of the results can be difficult to determine i.e. what percentage of the population does the simulation represent? You can get data overload if a simple figure of merit is not defined.